Media
2 items
Description
HallGuard is a local-first AI permission firewall that warns before risky prompts, uploads, and scam-like content. HallGuard: AI Security Firewall HallGuard is a local-first security layer that helps protect sensitive information before it is sent to AI websites. AI makes it easy to paste source code, credentials, customer information, confidential text, and other sensitive content into chat interfaces. HallGuard adds a safety checkpoint before those actions continue. What HallGuard protects against • Sensitive data Detect potentially sensitive information such as credentials, tokens, passwords, contact information, and other protected data patterns. • Prompt injection Identify suspicious instructions designed to manipulate AI systems or override intended behavior. • Scam and fraud signals Detect suspicious language associated with phishing, impersonation, urgency, payment requests, and social engineering. • Risky uploads Review file metadata such as filename, type, and size before an upload continues. Protection at the moment it matters HallGuard can inspect actions such as: • Pasting into AI chats • Sending or submitting messages • Pressing Enter to send • Clicking send controls • Selecting files for upload When a risk is detected, HallGuard can warn you, interrupt the action, or provide a redacted alternative depending on the detection and your settings. Privacy-first by design HallGuard performs sensitive-content analysis locally in the browser. Your full prompts are not sent to a remote ML inference service. Optional activity synchronization uses redacted information rather than storing your full prompts. Improvement telemetry is a separate optional feature and is designed around bounded, content-free data. HallGuard also continues local protection when network connectivity is unavailable. Supported AI websites HallGuard initially supports major AI services including ChatGPT, Claude, and Gemini. Additional HTTPS domains can be configured through HallGuard's protected-site controls. Built for transparency HallGuard uses layered detection combining deterministic security rules, heuristics, local classification, and policy decisions. No security system can detect every possible sensitive value or malicious instruction. HallGuard is designed as a preventive safety layer, not a guarantee that all threats or data leaks will be detected. Learn more about HallGuard, its privacy model, detection architecture, and limitations on the HallGuard website.